OpenMx Crash - double free or corruption - CT-SEM

2 replies [Last post]
rabil's picture
Offline
Joined: 01/14/2010

I'm trying to run the CT-SEM examples from Oud et al. I'm running on an 8-core system that's running Red Hat EL5. I compiled R locally (2.15.2 Patched (2013-01-22 r61734)) and I'm running OpenMx 1.3.2. I've attached the CT-SEM files (containing data, input, and continuous_time function) and the output when R crashes.

I've tried running the same code in the same manner on a dual-core laptop running OpenMx 1.3.2 and R version 2.15.1 (Ubuntu 11.10). R doesn't crash but the model never terminates even after 4 days of running. I can't tell what OpenMx is doing. According to the CT-SEM docs the code should take maybe 20 minutes to terminate.

Any ideas as to what is going wrong? Thanks.

AttachmentSize
openmx_crash.txt3.97 KB
BMSP_2043_sm_input1_rab_v1.R4.55 KB
BMSP_2043_sm_input3.R12.01 KB
BMSP_2043_sm_data1.dat85.11 KB
mhunter's picture
Offline
Joined: 07/31/2009
The code runs fine on my

The code runs fine on my single-threaded install of OpenMx. It only took about 5 minutes to estimate. I think the issue is with parallelism. While we try to find out what the issue is you could try installing OpenMx single-threaded on the 8-core machine.

The function(s) by Oud et al. are bending a lot of the RAM spec to their purposes (which is one of the great things about OpenMx in my opinion), but its wreaking some havoc internally.

Incidentally, I've been working on additions to OpenMx to estimate models like this as a built-in feature. Matrix specification of LISREL matrices currently works, and state space models for discrete time are also working (though not heavily tested). If I understand correctly, the function(s) by Oud et al. are doing state space models for continuous time.

rabil's picture
Offline
Joined: 01/14/2010
Single Thread OpenMx Solves the Problem

Thanks. I re-installed OpenMx and set it to single thread. The code runs and terminates in 2.75 minutes. I haven't tried it yet on my laptop.

I had to make some simple changes to the code so that it would store the results. As written, sourcing the files results in nothing printed or saved.

I'm definitely interested in using your discrete state space code.

Thanks again.